Prediction of protein-protein interactions from residue-pair interaction prediction

已经开发了用于预测蛋白质-蛋白质相互作用位点、相互作用伴侣和用于定义蛋白质-蛋白质复合物之间的相似性以用于分类的方法。核心方法是基于神经网络集成。通过在查询特定数据集上对模型进行实时训练，模型的准确性得到了显着提高。由于训练数据中蛋白质Q的存在而导致的蛋白质P的预测精度的提高被用作成对相似性度量。该模型的一个更快的版本被开发用于基因组规模的应用。

Methods for prediction of protein-protein interaction sites, interacting partners and for defining similarity between protein-protein complexes for classification have been developed. Core method is based on a neural network ensemble. Accuracy of the model is significantly improved by real-time training of models over query-specific data sets. Improvement in prediction accuracy of a protein P due to the presence of a protein Q in the training data is used as pair-wise similarity metric. A faster version of the model is developed for genomic scale application.